signed distances are not calculated if distance is >1e5

[jefferis]

I am working with some electron microscopy data calibrated in nm. I am using signed distance with vcgClostKD to determine if points are inside a mesh. Some of my distances from points outside the mesh are greater than 1E5.

Unfortunately 1E5 is being used as a signalling value in some contexts to mean out of range. The lines below effectively assumes that all distances are <1E5.

Rvcg/src/RvcgKD.h

Lines 188 to 193 in d689603

	if (sign && (distances[i] < 1e5)) {
	Point3f dif = clost - currp;
	//float sign = dif.dot(vertexnormal);
	if (dif.dot(vertexnormal) < 0)
	distances[i] = -distances[i];
	}

would it perhaps be possible to use a NAN/Inf or other signalling procedure?

[zarquon42b]

Hi Jeff,
I will look into this.

[zarquon42b]

Fixed in 63329d0: Set max dist to 1e12 and all beyond to NaN.
Please test and close issue if you deem it fixed.

[jefferis]

Thanks a lot. I'll take a look later but based on the code change I think this will work for me.

[jefferis]

Thanks again!

[jefferis]

PS @zarquon42b do you have any plans for a CRAN release in the near future?

[zarquon42b]

If you need it, I will make one. No big deal.

[jefferis]

Thank you! Would be very handy as Rvcg is a dependency for my packages and most of my users are on mac and do not have a development environment set up but can use binary packages from CRAN.

[zarquon42b]

Hi Greg,
FYI: it is now on CRAN
Best
Stefan